我正在做家庭作业，我一直被困扰着几个小时，一直在寻找如何为神经网络随机数生成的权重正确设置尺寸。无论我阅读了多少篇文章，还是我在Google搜索上发表的文章，都找不到解决方案。每次更改尺寸时，基于传入X_train集的尺寸，该程序最终都会出现“ ValueError：操作数无法与形状（X，X）（y，y）一起广播”的情况。主要问题是点积对2d数组进行数学运算的复杂方式。我不知道该去哪里，所以我在这里。我将发布成本和示例输出以提供尽可能多的信息，并每小时检查一次，以查看是否有人可以解决此问题。我真正需要的是一种千篇一律的方式说...如果您要推送一个n维数组，则权重应为这些维，权重应为这些维，权重为3，依此类推...依此类推不是兼容性错误。
我尝试在互联网上寻找一种基于数据结构传入维度来解密权重的方法。即行和列。
import numpy as np
import pandas as pd
from numpy import tanh
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
class NeuralNetwork():
def __init__(self):
print('constructor working...')
self.inputsize = 4
self.outputsize = 4
self.hiddensize = 1
self.W1 = np.random.randn(self.inputsize, self.hiddensize)
self.W2 = np.random.randn(self.hiddensize, self.outputsize)
def forward(self, X):
#print('forward - X:\n', X)
self.z = np.dot(X, self.W1)
#print('forward - self.z:\n', self.z)
self.z2 = self.sigmoid(self.z)
#print('forward - self.z2:\n', self.z2)
self.z3 = np.dot(self.z2, self.W2)
#print('forward - self.z3:\n', self.z3)
o = self.sigmoid(self.z3)
print('forward - o:\n', o)
print('forward shape of o:\n', o.shape)
print('forward shape of X:\n', X.shape)
return o
def sigmoid(self, s):
#print('sigmoid:\n', (1/(1+np.exp(-s))))
return(1/(1+np.exp(-s)))
def sigmoidPrime(self, s):
return(s * (1 - s))
def backward(self, X, y, o):
print('backward - X:\n',X,'\ny:\n',y,'\no:\n',o)
self.o_error = y - o
print('backward - o_error:\n', self.o_error)
self.o_delta = self.o_error * self.sigmoidPrime(o)
print('backward - o_delta:\n', self.o_delta)
self.z2_error = self.o_delta.dot(self.W2.T)
print('backward - z2_error:\n', self.z2_error)
self.z2_delta = self.z2_error * self.sigmoidPrime(self.z2)
print('backward - z2_delta:\n', self.z2_delta)
self.W1 += X.T.dot(self.z2_delta)
print('backward - W1:\n', self.W1)
self.W2 += self.z2.T.dot(self.o_delta)
print('backward - W2:\n', self.W2)
def train(self, X, y):
o = self.forward(X)
self.backward(X, y, o)
def saveWeights(self):
np.savetxt('w1.txt', self.W1, fmt='%s')
np.savetxt('w2.txt', self.W2, fmt='%s')
def predict(self):
print("Predicted data based on trained weights: ")
print("Input (scaled): \n" + str(X_test))
print("Output: \n" + str(self.forward(X_test)))
if __name__ == "__main__":
nn = NeuralNetwork()
titanic_original_df = pd.read_csv(r'./titanic_data.csv')
titanic_df = titanic_original_df.copy()
print('titanic data shape:', titanic_df.shape)
#print('titanic data head:\n', titanic_df.head(3))
'''
for col in titanic_df:
print(col,': ',titanic_df[col].dtypes)
for col in titanic_df:
print(col,'- value counts:\n', titanic_df[col].value_counts())
for col in titanic_df:
print(col,'- null data:', titanic_df[col].isnull().sum())
'''
titanic_df['Age'] = titanic_df['Age'].interpolate().round()
#print('after interpolation, Age null counts:\n', titanic_df['Age'].isnull().sum())
titanic_df['Sex'] = pd.get_dummies(titanic_df['Sex'])
#print('after dummy encoding Sex:\n', titanic_df['Sex'].value_counts())
for col in titanic_df:
print(col,'- null data:', titanic_df[col].dtypes)
titanic_df[['Pclass','Sex']] = titanic_df[['Pclass','Sex']].astype(np.float64)
sc = StandardScaler()
#scaled_data = sc.fit(titanic_df[['Age','Fare']])
#titanic_df[['Age','Fare']] = sc.transform(titanic_df[['Age','Fare']])
#print('after scaling, Age column:\n', titanic_df['Age'].value_counts())
#print('after scaling, Fare column:\n', titanic_df['Fare'].value_counts())
y = titanic_df.Survived
X = titanic_df.drop(['PassengerId','Survived','Name','SibSp','Parch','Ticket','Cabin','Embarked'], axis=1)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=124)
print('X_train shape:', X_train.shape)
print('X_test shape:', X_test.shape)
print('y_train shape:', y_train.shape)
print('y_test shape:', y_test.shape)
#print('X_train:\n', X_train['Sex'].value_counts())
for i in range(1):
print('# '+str(i)+'\n')
#print('Input (scaled):\n'+str(X_train))
#print('Actual output:\n'+str(y_train))
print('Predicted output:\n'+str(nn.forward(X_train)))
print('shape of X_train:',X_train.shape)
print('shape of y_train:',y_train.shape)
print('Loss:\n'+str(np.mean(np.square(y_train - nn.forward(X_train)))))
print('\n')
nn.train(X_train, y_train)
nn.saveWeights()
nn.predict()
In [55]: runfile('C:/Users/John/.spyder-py3/ProgrammingAssignment#9.py', wdir='C:/Users/John/.spyder-py3')
constructor working...
titanic data shape: (891, 12)
PassengerId - null data: int64
Survived - null data: int64
Pclass - null data: int64
Name - null data: object
Sex - null data: uint8
Age - null data: float64
SibSp - null data: int64
Parch - null data: int64
Ticket - null data: object
Fare - null data: float64
Cabin - null data: object
Embarked - null data: object
X_train shape: (623, 4)
X_test shape: (268, 4)
y_train shape: (623,)
y_test shape: (268,)
# 0
forward - o:
[[0.50384373 0.4961504 0.50183024 0.49790133]
[0.5001908 0.49980891 0.50009085 0.49989583]
[0.51753819 0.48243502 0.50835355 0.49042155]
...
[0.51554828 0.48442797 0.50740524 0.49150886]
[0.50025489 0.49974472 0.50012137 0.49986083]
[0.50000075 0.49999925 0.50000036 0.49999959]]
forward shape of o:
(623, 4)
forward shape of X:
(623, 4)
Predicted output:
[[0.50384373 0.4961504 0.50183024 0.49790133]
[0.5001908 0.49980891 0.50009085 0.49989583]
[0.51753819 0.48243502 0.50835355 0.49042155]
...
[0.51554828 0.48442797 0.50740524 0.49150886]
[0.50025489 0.49974472 0.50012137 0.49986083]
[0.50000075 0.49999925 0.50000036 0.49999959]]
shape of X_train: (623, 4)
shape of y_train: (623,)
forward - o:
[[0.50384373 0.4961504 0.50183024 0.49790133]
[0.5001908 0.49980891 0.50009085 0.49989583]
[0.51753819 0.48243502 0.50835355 0.49042155]
...
[0.51554828 0.48442797 0.50740524 0.49150886]
[0.50025489 0.49974472 0.50012137 0.49986083]
[0.50000075 0.49999925 0.50000036 0.49999959]]
forward shape of o:
(623, 4)
forward shape of X:
(623, 4)
Traceback (most recent call last):
File "<ipython-input-55-52d7c067a2dd>", line 1, in <module>
runfile('C:/Users/John/.spyder-py3/ProgrammingAssignment#9.py', wdir='C:/Users/John/.spyder-py3')
File "C:\Users\John\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 786, in runfile
execfile(filename, namespace)
File "C:\Users\John\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 110, in execfile
exec(compile(f.read(), filename, 'exec'), namespace)
File "C:/Users/John/.spyder-py3/ProgrammingAssignment#9.py", line 117, in <module>
print('Loss:\n'+str(np.mean(np.square(y_train - nn.forward(X_train)))))
File "C:\Users\John\Anaconda3\lib\site-packages\pandas\core\ops.py", line 1583, in wrapper
result = safe_na_op(lvalues, rvalues)
File "C:\Users\John\Anaconda3\lib\site-packages\pandas\core\ops.py", line 1529, in safe_na_op
return na_op(lvalues, rvalues)
File "C:\Users\John\Anaconda3\lib\site-packages\pandas\core\ops.py", line 1505, in na_op
result = expressions.evaluate(op, str_rep, x, y, **eval_kwargs)
File "C:\Users\John\Anaconda3\lib\site-packages\pandas\core\computation\expressions.py", line 208, in evaluate
return _evaluate(op, op_str, a, b, **eval_kwargs)
File "C:\Users\John\Anaconda3\lib\site-packages\pandas\core\computation\expressions.py", line 123, in _evaluate_numexpr
result = _evaluate_standard(op, op_str, a, b)
File "C:\Users\John\Anaconda3\lib\site-packages\pandas\core\computation\expressions.py", line 68, in _evaluate_standard
return op(a, b)
ValueError: operands could not be broadcast together with shapes (623,) (623,4)
答案 0 :(得分：2)
在Python编程中不必花太多时间。 转发数据的主要问题是权重必须“适合”神经元。
在数学表达式中，您可以说一个简单的点积，例如： [2 x 3]矩阵点乘[3 x 1]产生[2 x 1]矩阵，请注意，矩阵乘法的方向很重要。
然后您可以将其拆分为矩阵A，其中n行和m列点乘以矩阵B，矩阵B必须具有m列（!!）和可选的列数，比方说z列。结果A x B >>导致矩阵C的形状为[n x z]。
在代码中，您可能会遇到数组大小的错字，缺少转置等的情况。